From Pointwise Bounds to a Baire-Category Argument
The Uniform Boundedness Principle, also called the Banach–Steinhaus Theorem, begins with a bound for each fixed input vector and concludes with one bound for an entire family of operator norms. The central proof question is how to turn the separate bounds into a common one. The proof architecture has three parts: encode the pointwise bounds as closed sets, use completeness to find one such set with interior, and convert its interior into a bound on operator norms.
Let \(X\) be a Banach space, let \(Y\) be a normed space, and let \(\mathcal{T}\) be a family of bounded linear operators from \(X\) to \(Y\). For a nonnegative integer \(m\), consider the set of points where every operator in the family takes a value of norm at most \(m\). These sets organize the pointwise hypothesis into a countable cover of \(X\).
The sets are nested: \(E_m\subseteq E_{m+1}\). More importantly, pointwise boundedness says that every \(x\in X\) belongs to at least one of them. For a fixed \(x\), the supremum of \(\|Tx\|\) over \(T\in\mathcal{T}\) is finite, so some integer \(m\) is at least that supremum. Thus \(X=\bigcup_{m=0}^{\infty}E_m\). The Baire Category Theorem becomes relevant only after we verify that these level sets are closed.
Proof. A bounded linear operator is continuous. Consequently, for each \(T\in\mathcal{T}\), the map \(x\mapsto\|Tx\|\) is continuous: if \(x_j\to x\), then \(\bigl|\|Tx_j\|-\|Tx\|\bigr|\leq\|T(x_j-x)\|\leq\|T\|\|x_j-x\|\to0\). Therefore the set \(\{x:\|Tx\|\leq m\}\) is closed, as the inverse image of the closed interval \([0,m]\) under a continuous real-valued function. Since an arbitrary intersection of closed sets is closed, the defining intersection for \(E_m\) is closed. \(\square\)
The intersection may run over an uncountable family; closedness does not require the family to be countable. Countability enters elsewhere: the pointwise bounds are gathered into the countable collection \(E_0,E_1,\ldots\), which covers the space. This distinction is useful when reading the proof. Baire category is applied to the countable cover by level sets, not to a listing of the operators.
The Proof Certificate
The following result makes the proof architecture quantitative. It records not only that the operator norms are bounded, but also how a particular level set and an open ball inside it provide a numerical bound. This is the Baire-category proof mechanism for the Uniform Boundedness Principle recalled in the previous tutorial.
Proof. The level sets \(E_m\) are closed by the lemma, and pointwise boundedness gives \(X=\bigcup_{m=0}^{\infty}E_m\). The Baire Category Theorem says that a complete metric space cannot be a countable union of closed sets all having empty interior. Since \(X\) is complete, at least one \(E_m\) has nonempty interior. Hence there are \(x_0\in X\) and \(r>0\) such that the open ball \(B(x_0,r)\) is contained in \(E_m\).
The proposition “Local Bound on a Ball Controls Operator Norms,” established in the previous tutorial, now applies. Every \(x\in B(x_0,r)\) satisfies \(\|Tx\|\leq m\) for every \(T\in\mathcal{T}\), so the proposition gives \(\sup_{T\in\mathcal{T}}\|T\|\leq2m/r\). This is finite because \(m\) is an integer and \(r\) is positive. The bound is independent of the operator, which is the desired uniform conclusion. \(\square\)
The proof has a useful division of labor. Pointwise boundedness provides the cover, continuity provides closedness, and completeness supplies a level set with interior. Linearity is essential in the ball-to-norm step: it lets a bound on a ball control the changes \(Th\), and hence the action on every direction. No one step alone gives the conclusion.
Worked Example: Integral Functionals and Their Level Sets
Let \(X=C([0,1])\) with the supremum norm, and define \(T_n:X\to\mathbb{R}\) by \(T_n(f)=\int_0^1 t^n f(t)\,dt\), for \(n=0,1,2,\ldots\). Each \(T_n\) is linear. Also, \[ |T_n(f)|\leq\int_0^1 t^n|f(t)|\,dt \leq\|f\|_\infty\int_0^1t^n\,dt =\frac{\|f\|_\infty}{n+1}. \] The last equality follows by integrating \(t^n\) on \([0,1]\). In particular, \(T_n\) is bounded and \(|T_n(f)|\leq\|f\|_\infty\) for every \(n\). Thus for each fixed \(f\), the pointwise supremum is finite, and any integer \(m\geq\|f\|_\infty\) satisfies \(f\in E_m\).
The norm of each functional can also be calculated exactly. The displayed estimate gives \(\|T_n\|\leq1/(n+1)\), while the constant function \(f(t)=1\) has norm \(1\) and satisfies \(T_n(f)=1/(n+1)\). Hence \(\|T_n\|=1/(n+1)\). This example displays the level-set construction explicitly: all the functionals are controlled at each fixed input, and in fact their norms already have a common bound. The Baire argument is designed to obtain uniform boundedness even when such a direct estimate is unavailable.
Worked Example: An Uncountable Family on \(\ell^1\)
Let \(X=\ell^1\), and let \(\mathcal{T}\) consist of the functionals \(T_\sigma(x)=\sum_{k=1}^{\infty}\sigma_kx_k\), where each \(\sigma=(\sigma_k)\) has \(\sigma_k\in\{-1,1\}\). The series converges absolutely because \(\sum_{k=1}^{\infty}|\sigma_kx_k|=\sum_{k=1}^{\infty}|x_k|=\|x\|_1\). Each \(T_\sigma\) is linear and satisfies \(|T_\sigma(x)|\leq\|x\|_1\), so it is bounded with \(\|T_\sigma\|\leq1\). Taking \(x=e_1=(1,0,0,\ldots)\) gives \(\|e_1\|_1=1\) and \(|T_\sigma(e_1)|=1\); therefore \(\|T_\sigma\|=1\).
For a fixed \(x\in\ell^1\), choose \(\sigma_k=1\) when \(x_k\geq0\) and \(\sigma_k=-1\) when \(x_k<0\). Then \(\sigma_kx_k=|x_k|\) for every \(k\), so \(\sup_\sigma|T_\sigma(x)|=\|x\|_1\). It follows that the level set \(E_m\) is exactly the closed ball \(\{x:\|x\|_1\leq m\}\). Indeed, the supremum formula shows that membership in \(E_m\) is equivalent to \(\|x\|_1\leq m\). There are uncountably many sign sequences, but the proof architecture still applies: the level sets are closed and form a countable cover.
Why Completeness Is Not a Technical Decoration
Completeness is what allows the Baire Category Theorem to turn a countable closed cover into interior for at least one member of the cover. Without completeness, pointwise boundedness need not imply uniformly bounded operator norms. The next example exhibits the failure in a familiar normed space that is not complete.
Worked Example: Pointwise Bounded Operators with Unbounded Norms on \(c_{00}\)
Let \(c_{00}\) be the vector space of real sequences with finite support, equipped with the \(\ell^2\) norm. For each positive integer \(n\), define \(T_n:c_{00}\to\mathbb{R}\) by \(T_n(x)=n x_n\). This is linear, and \[ |T_n(x)|=n|x_n|\leq n\left(\sum_{k=1}^{\infty}|x_k|^2\right)^{1/2} =n\|x\|_2. \] Thus \(T_n\) is bounded and \(\|T_n\|\leq n\). For the unit vector \(e_n\), \(\|e_n\|_2=1\) and \(T_n(e_n)=n\), so \(\|T_n\|\geq n\). Therefore \(\|T_n\|=n\), and the operator norms are unbounded.
Nevertheless, the family is pointwise bounded. For any fixed \(x\in c_{00}\), only finitely many coordinates are nonzero, so the sequence of values \(T_n(x)=nx_n\) is zero except at finitely many indices. A sequence with only finitely many nonzero real values has a finite supremum of its absolute values. Thus \(\sup_n|T_n(x)|<\infty\) for every \(x\in c_{00}\).
The level sets make the breakdown visible. They cover \(c_{00}\) by pointwise boundedness and are closed by the level-set lemma. But none has interior. To verify this, fix \(m\geq0\), \(x\in E_m\), and \(\varepsilon>0\). Since \(x\) has finite support, choose \(n\) outside that support so large that \(n\varepsilon/2>m\). Set \(y=x+(\varepsilon/2)e_n\). Then \(\|y-x\|_2=\varepsilon/2<\varepsilon\), while \(|T_n(y)|=n\varepsilon/2>m\), so \(y\notin E_m\). Hence no open ball around any point of \(E_m\) is contained in \(E_m\). The Baire step fails because \(c_{00}\) is not complete.
Reading the Architecture Without Overclaiming
The proof does not find a bound by taking the largest of the pointwise bounds; there may be no such largest bound, and each bound is attached to a different input. Instead, the countable level sets turn those varying bounds into a cover of the domain. Baire category forces one level set to be large enough to contain a ball, and linearity converts that local information into a bound valid for every direction. The constant obtained from the ball, \(2m/r\), is a certificate arising from this particular proof; it need not be the smallest possible uniform bound.
When applying the argument, check each part separately: the domain is complete, the operators are bounded and linear, and the family is pointwise bounded. Then form the level sets and use the Baire Category Theorem. If the domain is incomplete, as in the \(c_{00}\) example, the pointwise hypothesis alone is not enough. If the maps are not linear, a ball bound need not control their behavior at arbitrary scales. The architecture identifies not just the conclusion, but exactly where its assumptions do their work.
Check Your Understanding
Use the level-set proof architecture to answer the following questions.
- Why is each uniform-bound level set closed, even when the operator family is uncountable?
- Which countable collection of sets covers the domain, and which hypothesis guarantees the cover?
- What does the Baire Category Theorem provide that is needed to bound operator norms?
- Where does linearity enter when passing from a bound on a ball to a bound on operator norms?
- In the \(c_{00}\) example, why can a small perturbation create a value of \(T_n\) larger than a fixed level \(m\)?